Method and apparatus for verifying data local to a single thread

Abstract

Concurrent program analysis is fashioned for detecting potential race conditions such as data races in computer programs. One feature of this analysis is verifying annotations of addressable resources in a program. Annotations are verified by checking if thread-local resources are indeed thread-local, and that thread-shared data spaces are not in fact thread-local. Another feature is detecting potential race conditions, such as data races, in the computer program. The computer program can spawn a plurality of threads that are capable of being executed concurrently. The source code of the computer program being analyzed includes an element annotated as either thread-local or thread-shared. The validity of the thread-local annotation is verified if the element is annotated in the computer program as thread-local, wherein an invalid thread-local annotation may cause a race condition.

Description

1. Field of InventionThis invention relates generally to concurrent program analysis and more specifically to tools for detecting races such as data races in multithreaded programming.2. Background of the InventionConcurrent programming involves multiple concurrent flows of control to perform multiple tasks in parallel. Each time a task is invoked, a thread is used to implement the particular task. A "thread" is a single sequential flow of control that can be abstractly thought of as an independent program control block. Each thread can execute its instructions independently, allowing a multithreaded program to perform numerous tasks concurrently. Having multiple threads in a program means that multiple threads can be executing in parallel, where each thread is running its own execution path. Namely, at any instance the multithreaded program has multiple points of execution, one in each of its threads. Additionally, a multithreaded program creates a new thread by some mechanism, suc...

AI Technical Summary

Benefits of technology

In one embodiment of the present invention, checks are run on elements of a computer program source code that are annotated as thread local in order to determine the validity of their respective thread-local annotations. Alternatively stated, checks are run on the elements of the source code that are annotated as thread-local to verify that such elements are indeed thread-local and not in fact thread-shared. Further examination can determine if elements of the source code that are annotated as thread-shared are in fact sharable. Additional requirements are needed to ensure the validity of the thread-local annotations when one addressable resource is derived from another as when one class is derived from another. Programs that satisfy these requirements are guaranteed not to share values of thread-local type between multiple threads. Therefore, the analysis enables static tools to properly check programmer annotations indicating which data types are thread-shared and which are thread-local.

Problems solved by technology

In multithreaded programming, access of thread-shared addressable resources requires special attention since multiple threads can operate in parallel to manipulate them and, as a result, errors may arise.

Race conditions often result in non-deterministic program behavior.

Escape analysis is not capable of verifying thread-local annotations if only part of the program is provided.

Inferring thread-local properties requires substantially more computational resources than just verifying them and therefore is not as efficient.

Conventional static analysis tools, such the Warlock, are not capable of verifying that certain data locations have the thread-local property.

However, Warlock does not verify whether the thread-local annotation is correct.

An invalid thread-local annotation may cause a race condition.

Images